Addressing flux suppression, radio frequency interference, and selection of optimal solution intervals during radio interferometric calibration

Abstract

The forthcoming Square Kilometre Array is expected to provide answers to some of the most intriguing questions about our Universe. However, as it is already noticeable from MeerKAT and other precursors, the amounts of data produced by these new instruments are significantly challenging to calibrate and image. Calibration of radio interferometric data is usually biased by incomplete sky models and radio frequency interference (RFI) resulting in calibration artefacts that limit the dynamic range and image fidelity of the resulting images. One of the most noticeable of these artefacts is the formation of spurious sources which causes suppression of real emissions. Fortunately, it has been shown that calibration algorithms employing heavy-tailed likelihood functions are less susceptible to this due to their robustness against outliers. Leveraging on recent developments in the field of complex optimisation, we implement a robust calibration algorithm using a Student's t likelihood function and Wirtinger derivatives. The new algorithm, dubbed the robust solver, is incorporated as a subroutine into the newly released calibration software package CubiCal. We perform statistical analysis on the distribution of visibilities and provide an insight into the functioning of the robust solver and describe different scenarios where it will improve calibration. We use simulations to show that the robust solver effectively reduces the amount of flux suppressed from unmodelled sources both in direction independent and direction dependent calibration. Furthermore, the robust solver is shown to successfully mitigate the effects of low-level RFI when applied to a simulated and a real VLA dataset. Finally, we demonstrate that there are close links between the amount of flux suppressed from sources, the effects of the RFI and the employed solution interval during radio interferometric calibration. Hence, we investigate the effects of solution intervals and the different factors to consider in order to select adequate solution intervals. Furthermore, we propose a practical brute force method for selecting optimal solution intervals. The proposed method is successfully applied to a VLA dataset.